I originally considered computing cast shadow patterns based on DEM obtained from SRTM. Since all my datasets were in non-projected WGS84 reference coordinates, I applied r.sunmask to the DEM datasets. I observed that the resulting cast shadow exhibited a pattern apparently independent on sun elevation, an unexpected result requiring more investigation to understand. Back to basics, I created a very basic elevation model (different headers for projected -- WGS84/UTM31N or non-projected -- WGS84 frameworks when loading in GRASS) using GNU/Octave in which a trapezoidal mountain is located in a flat plain at elevation 0 (less beautiful than the gaussian volcano of the R.sun tutorial but easier to generate). Notice that the 0-elevation baseline requires to NOT ignore this altitude in the r.sunmask processing. Performing the same experiment (cast shadow calculation) using this basic dataset, I indeed confirmed that the shadow length was independent on sun elevation when using an unprojected reference framework.

Left (click on image for a full resolution version): Digital Elevation Model -- Right: unprojected framwork for various sun elevations: the cast shadow length is independent on sun elevation (45 and 80 degree elevations) !

Based on this analysis, I considered using a projected framework, namely WGS84/UTM31N (instead of the basic unprojected WGS84, since I am interested in analyzed time-lapse pictures grabbed in the French Alps). Before investigating the real (and computationaly intensive) DEM from the Alps (as provided by the SRTM), I used the basic model on a projected framework to indeed verify that the cast shadow is dependent on sun elevation in r.sunmask.

Thess images were obtained by varying the elevation for each new azimuth of the sun position (from 20 to 45 degrees for azimuth from 225 to 45 degs). Based on these results, it was concluded that a projected reference framework was needed: for France, WGS84/UTM31N. After loading the WGS84-referenced SRTM DEM in QGis, the file was saved in a new projected framework WGS84/UTM31N before being loaded into GRASS in a map which is referenced to WGS84/UTM31N. Once this is done, the dataset is restricted to the GRASS region using r.resample (don't forget to overload the projection reference when processing the file). The result is as follows, including a geometrically corrected picture from the webcam located at Aiguille du Midi (Mont Blanc area near Chamonix, France).

Once the restricted DEM is selected various configurations of sun position are considered to compute the cast shadow. Most significantly, one picture attracted our attention (Sept. 22 2013 at 13h05 local time) in exhibiting the cast shadow of two mountains.

These results are well visible in the following cast shadow calculations. Notice that the image properties are displayed since using these cast shadow masks is not obvious at first sight: I select a transparency mask based on the calculation result (NaN generates 100% transparency) and a background color with a transparency of 65% in order to observe the background image.
First the cast shadow map on a slope map:

and the shadow map displayed over a backgroud image geometrically corrected to match the topography:

Above is the computation for an elevation of 44.2 degrees and an azimuth of 174.7 degrees as provided on the USNO web site for the sun position at the time the picture was taken by the webcam (13/09/22 at 1305). The cast shadow of the mountain is well visible over the glacier, although a detailed analysis beyond the visual comparison has not been performed yet.